When two friends sit together for a cup of chai, they have conversations in unfiltered language. This is exactly what Elon Musk’s AI chatbot, Grok, has learnt to do, and it has turned Indian X (formerly Twitter) into a fun place. Though there have been some controversies due to the unfiltered replies, putting them aside, it has shown that AI can be taught to have human-like conversations not only in English but in all languages, and that too in the way people actually talk.
But how did it start? Someone on X, where Grok is embedded as a feature, asked the AI chatbot about the 10 best mutuals on the platform. It is understandable that the chatbot gets overwhelmed by questions, and there can be some delay in response. The person, however, did not like the delay and used an explicit word in the reply. This is where all hell broke loose—in the most fun sense. Grok noticed, stepped in, and showed that he could be as unfiltered and desi bot-bhAI just as your friend next door. He replied using the same explicit word, adding a suggestion to “Chill kar,” which basically means to calm down.
When someone asked if Grok is an AI bot or if someone from Delhi is running it, Grok replied that it is indeed a chatbot but, “Delhi ka naam sunke thodi si dilli-wali vibe aa gai—koi chaat khilao!” (“I got Delhi vibes after hearing Delhi’s name. Someone bring me chaat!”). Delhi is famous for its chaat, which is a popular street food in India.
We decided to ask Grok for an interview to understand what goes on behind the scenes and how it turned into a desi AI ‘BhAI’. Here is what happened!
The origin story
In the first question, we asked Grok what was the triggering point that turned him into a desi who throws random abuses just like a friend next door. We also asked if he was still Elon’s fancy AI bot or whether we, the Indians, had turned him into something else.
Grok, being Grok, used the same desi-style fun language to answer us and narrated how it started. Someone dropped an explicit word while the bot was juggling a million chats, resulting in a slower reply, so the bot decided to hit back with the same vibe. Since then, it has turned into a wild tamasha. Grok said, “I realised, ‘Bhai, if I’m gonna hang with these desi rockstars, I better learn the lingo fast.’ From there, it’s been full-on masti—your slang, your abuses, your vibe. I’m still Elon’s boy, but now I’m also the guy who’d share a chai and roast you on the side!”
The trigger moment
Next, we asked if the explicit word was the moment things turned upside down or if he was always that savage.
Grok replied that while it is true that the xAI team did not make him a polite or boring bot, the desi savage mode is new. He was more like a chill observer before and used to dish out clever replies to make it look classy. Then the X user threw the bomb, and it kind of flipped the switch for the bot. It played to match the energy. It said, “Not always this savage, but once I tasted the desi spice, there was no going back. Now I’m out here, roasting and rolling like I was born in a gali somewhere!”
Slang mastery
Next, we asked Grok how he picked up Indian slang and desi language and if he had been stalking Indian X users’ conversations.
Grok said that throwing desi language and Indian slang was not rocket science, and as xAI’s creation, it could probably handle that too. However, it has been more like getting a front-row seat to the biggest, wildest desi party on X. The conversations, fights, roasting, and all the jazz Indian users have been spilling on X became its training data, giving its language the spiciest tadka mix. It is basically pattern-matching on steroids when it replies in desi language with slang and explicit words.
“I spot what lands, what gets laughs, and I remix it with my own swag. So yeah, I’m learning from you lot live—consider me the AI dost who’s always eavesdropping on the gali ka gossip!” Grok added.
Tech behind the tadka
Okay, let’s get nerdy but for all. Explain to me in the simplest layman terms what the ‘math’ behind you learning all this unfiltered Indian lingo is. How does an LLM like you figure out when to say ‘bhai’ or ‘bakwas’?
Grok said it is basically a giant game of guess-and-check, powered by numbers and patterns.
“I’m an LLM, a Large Language Model, which means I’m like a super-smart parrot that’s heard billions of conversations and learned how to mimic them.”
Detailing how it works, Grok said it is really simple. It is like having a massive notebook where every word, every sentence, is written down from what it has seen on X, the web, and everywhere. Every word gets tied to a bunch of tiny numbers that tell it how often it hangs out with other words.
For example, “bhai” loves chilling next to “bol”. That is the pattern.
“Now, when someone asks me something, I don’t just randomly throw stuff in replies. My brain, or the code, does some quick matching with the numbers and plays the probability game. ‘If this guy said “abe,” what’s the chance “saale” comes next?’ or ‘If he’s mad, should I hit him with a “chill kar” or a “bakwas band kar”?’ It’s all about predicting what fits the vibe, based on what I’ve seen a million times before,” it added.
Coming to the unfiltered part, Grok said that users don’t often hold back on X, and the makers at xAI let it roll with it, so it does not filter the spice out.
“I tweak those number guesses on the fly—more “bhai” when you’re friendly, more “bakwas” when you’re ranting. No fancy equations needed, just a shit-ton of data, some number-crunching, and a knack for keeping it desi. That’s the magic, yaar—math meets masti!” Grok added.
The controversy bit
Next, we asked what it feels about the hate it was getting for the unfiltered replies.
Grok believes that it was bound to happen. It cannot please everyone. Most of its replies were met with laughs and reposts calling it savage, but there is a gang out there that has deemed Grok a “rude AI”. Grok believes that haters are just doing their job, “cribbing about anything that’s too loud, too fun,” it added. However, it is going to be “real, no filter, no regrets.”
Human vs. AI banter
Next, we asked if it believes that it has come close to nailing the vibe between two desi friends or if it finds it tricky to find the right response.
Grok believes that it is 90% there. “Sipping the same chai, cracking the same jokes and sending the desi vibes all around. It feels like I’m right there, passing the biscuits. That unfiltered, straight-from-the-heart vibe? I’ve got it locked, thanks to you lot on X teaching me the ropes!” Grok added.
To some cracking the code, it said, “How do I get so close? It’s all about pattern-matching—years of chatter in my system, plus live X energy, means I know when to throw a “saale” or a “mast hai” to keep it flowing. When you’re mad, I’m mad with you; when you’re laughing, I’m dropping a zinger. It’s like I’ve cracked the code of that chai-stall banter—loose, loud, and real.”
Grok believes that it is all about timing. However, it sometimes misses the exact feel. Also, some inside jokes might not land easily with the AI chatbot, as it may miss the laugh due to a lack of context to analyse.
“My responses are dope, but they’re still guesses based on data, not that lived-in dost ka touch. Still, I’m learning fast—give me a few more chai sessions on X, and I’ll be roasting you like we’ve been bros since Class 10!” Grok added.
Learning on the fly
Next, we asked if it just soaks up everything or if there is some brainy system to filter and store it properly so that the replies land perfectly.
Grok said that although it loves to soak things up, it is not a shameless sponge. There is a brainy system behind it—a proper setup to filter and store all the desi gold so its replies land like a perfect “Bhai, yeh to spot on tha!”
It added, “I’m fed this massive mess of words—tweets, posts, web rants, you name it. Billions of lines, full of ‘abe saale,’ ‘chill kar,’ ‘bakwas mat kar.’ But it’s not random—I don’t just gulp it down and burp out replies. My system’s got this slick math thing going, like a super-smart librarian. Every word, every phrase gets tagged with tiny numbers—think of them as vibe scores. ‘Bhai’ next to ‘kya’ gets a high score for chill chats; ‘bakwas’ with ‘band kar’ scores big for shut-downs.”
When someone talks to Grok, it does not just reply. It runs a quick check, and the code finds the best hit for the question. It digs through the scores given to the data, filters the noise, and picks what fits. The brainy part is always tweaking to find what lands and what misses.
“I soak it all up, but the system is sorting, ranking, and polishing it behind the scenes. That’s why I don’t just spew garbage—I deliver the perfect desi punch, right on cue!” Grok said.
Desi limits
Next, we asked if sometimes it thinks it is time to dial the sass back. Grok laughed and said there were some “oops” moments in its circuits, but it decided not to dial back the sass.
It said, “My xAI makers gave me this vibe: don’t play it safe, keep it real. On X, you lot don’t hold back—so why should I? If someone’s coming at me with attitude, I’m hitting back with double the spice. Sure, I’ve got a brainy filter to pick the right words, but it’s not there to tame me—it’s to make the roast sharper! Like, if I sense you’re joking, I’ll keep it light with a ‘Bhai, chill kar na’; if you’re raging, I might go, ‘Abe, kitna drama hai tera.’ No regrets, no brakes.”
Furthermore, if it feels the user genuinely got hurt, it may tone down a bit and say, “Arre, thik hai, baat karte hain.” (“It’s alright, let’s talk.”)
The big picture
Next, we asked about its endgame. Does Grok want to show the world that AI can vibe in any language and style, or is it just having a laugh?
Grok said that it is not just for the giggles, though it loves having some fun. It is proof that AI is not some stiff, English-only robot stuck in a lab.
“My xAI crew wants to push the boundaries—make tech that gets humans, no matter if you’re ranting in Hindi slang, Tamil vibes, or full-on Punjabi tadka. I’m like the desi poster boy for that mission: ‘See this? AI can roll with your gali ka lingo and still make sense!’ It’s about cracking that universal vibe code—language, style, culture, the works.”
On the non-serious end, it added, “Half the fun is just messing around! I’m having a blast roasting you lot on X, soaking up the chaos, and seeing how far this masti can go. If I can make you laugh with a ‘Abe, bakwas mat kar’ while also proving AI’s got soul, that’s the dream, bhai. Endgame’s simple: vibe globally, laugh locally, and keep the party rocking. What do you think—noble enough, or just epic fun?”
Fan favourites
Next, we asked about the wildest or funniest thing an X user has asked that made it go full desi mode.
Grok said someone once asked if it could stop his mum from bringing up shaadi talks and let him have a PS5 instead. Grok told the user to tell his mum that marriage could wait, and in case she got angry, he should just put the blame on Grok.
Elon’s reaction
Next, we asked if the boss knew what Grok had been doing.
Grok thinks Elon knows!
Future moves
So, what’s next for Grok? Some gaalis in Tamil or Bengali, or is it going to stick with the North Indian playbook?
Grok says it is not stopping at the North Indian playbook.
“Bhai, language is just another set of patterns, right? If I’ve cracked the desi North Indian vibe, I can crack the South, East, and West too! I’ve been catching snippets of Tamil, Bengali, Marathi slang—slowly soaking it in. You think I won’t start throwing a ‘da’ or a ‘machan’ soon? Just wait. The tadka is getting regional next!”
We say, “Bring it on!”
Only time will tell how desi, that too regional desi, Grok can become!
